SPORADIC: An Anytime Architecture for Intermittent Computing
Karthik Ganesan · TSpace (University of Toronto) · 2018
Energy-harvesting is an emerging computing paradigm that allows systems to run solely off energy harvested from their environment. However, these systems are subject to frequent and unpredictable power outages. Ensuring forward progress under such circumstances is crucial. Energy-harvesting applications are typically amenable to approximation. We propose SPORADIC, a set of anytime approximation techniques: Subword pipelining, Subword vectorization and Skim points. Subword pipelining and Subword vectorization decompose data words so that they can be processed as subwords, sequentially from most to least significant. Skim points fundamentally decouple the checkpoint location from the recovery location upon a power outage. SPORADIC transforms the processing on energy-harvesting devices from all-or-nothing to as-is computing. We find that under frequent power outages, SPORADIC yields speedups of 2.16× and 3.02× on non-volatile and checkpoint-based volatile processors, while still producing high-quality outputs.